INDEX

Explanations

terms related to falsehoods or deceiving statements

oai_token-act-pair · gpt-3.5-turbo

occurrences of the word "lies."

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.0.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 zoom

-0.66

 Chapel

-0.63

 crisp

-0.61

 quint

-0.61

 better

-0.58

 curb

-0.58

 speed

-0.57

 dash

-0.56

 tattoo

-0.56

 Singer

-0.56

POSITIVE LOGITS

lies

5.06

lie

2.15

lied

1.86

lying

1.84

liest

1.59

pins

1.51

lier

1.43

liness

1.38

mares

1.30

li

1.26

Activations Density 0.008%

terms related to falsehoods or deceiving statements

occurrences of the word "lies."

No Comments

No Known Activations